Do you know on average our braincan only store SEVEN numbers at atime In a way the advance oftechnology is always at odds with theadvance of human evolution for ourcognitive abilities becomeincreasingly depleted with the onsetof data

How do we extract efficient information How do we predict and forecast Andhow do we visualise statistics to understand probabilities These are some ofthe fundamental questions that data science aims to tackle so that soon (wehope) more people are able to make their data make an impact

Strata Data ConferenceLondon 2018-- Catherine May 2018 This year Quantess London had theopportunity to participate andcontribute at the Strata DataConference in London Wesponsored our member Christinawhom some of you might rememberfrom our seminar last year topresent at the FinDay eventdiscussing her work on thefascinating topic of Sentiment

One of the main takeaways from theconference however should comewith a caution That is no matterwhat you are working on with thedata first keep it simple Most of us have by now heard ofdeep learning A multi-layeredneutral network that selects featuresand computes weights in a recursiveform is able to solve complex fluidproblems better than the classicalmethods we are more familiar withfrom econometrics Indeed there aremany benefits and good logic as towhy researchers are putting so much

Analysis Well done Christina We love to hear from members ofQuantess -- please continueto reach out to us for any ideas toshare or would like to take partin any of our future events ) If you ever think financial jargonswere confusing to keep up with whenyou first started in the City (as formost of us) well think again Techpeople love jargons as much as anyother trades - whats data wranglingApache data lakes Keras LSTMmodels and $parasectcent Over the three-day conference atStrata I had the opportunity toattend various workshops exposand keynote speeches It wasexciting to feel the pace and see howpeople are working very hard onthe data problem from all frontiers -whether it is parsing datasets (jargonI know) or building algos its neverstraightforward and often it requirespeople to be creative problemsolvers to frame the question andinvent tools For instance theproblem oftransforming unstructured unlabeleddata into something that is usablecan be very tricky and time-consuming especially if you aredealing with a scale that requiresterabytes to be processed The veryrecent work done by ICIJon investigative journalism is a greatcase on this

time and resources on deeplearning First it works much betterwith lots of data More importantlythe cost of trial and error has comedown significantly with the advanceof computational power (ever heardof parallel computing) But thedevils double comes in the form ofdata mining and overfitting When amodel no longer works often wehave little understanding or ability topick apart the complexity Itsimportant to combine thefundamentals with your algorithmsSo better keep it simple - throw a fewtricks and understand how to make itwork rather than go on a treasurehunt for the latest mathematicallydense paper That said here is a (short) paper foryou if you have time and want to feelvindicated that simple time-seriesmethods are still far superior thanwhat neutral network is attempting toestablish -- I invite you to readthe Uber study and you can drawnyour own conclusion )

Visit Quantess London

Upcoming Events- Quantess London July Seminar - We are hosting Quantess London Summer Seminar on 13th July Details of theevent and the speakers will be updated soon Stay tuned - Conferences on AI - Browse online for anything you are interested Quantess members are invited to attend this conference in June at a 20discount on current price (code SPK20)

Copyright copy 2018 Quantess London All rights reserved

Want to change how you receive these emails You can update your preferences or unsubscribe from this list